4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <asm/uaccess.h>
42 /* [Feb-1997 T. Schoebel-Theuer]
43 * Fundamental changes in the pathname lookup mechanisms (namei)
44 * were necessary because of omirr. The reason is that omirr needs
45 * to know the _real_ pathname, not the user-supplied one, in case
46 * of symlinks (and also when transname replacements occur).
48 * The new code replaces the old recursive symlink resolution with
49 * an iterative one (in case of non-nested symlink chains). It does
50 * this with calls to <fs>_follow_link().
51 * As a side effect, dir_namei(), _namei() and follow_link() are now
52 * replaced with a single function lookup_dentry() that can handle all
53 * the special cases of the former code.
55 * With the new dcache, the pathname is stored at each inode, at least as
56 * long as the refcount of the inode is positive. As a side effect, the
57 * size of the dcache depends on the inode cache and thus is dynamic.
59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60 * resolution to correspond with current state of the code.
62 * Note that the symlink resolution is not *completely* iterative.
63 * There is still a significant amount of tail- and mid- recursion in
64 * the algorithm. Also, note that <fs>_readlink() is not used in
65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
66 * may return different results than <fs>_follow_link(). Many virtual
67 * filesystems (including /proc) exhibit this behavior.
70 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
72 * and the name already exists in form of a symlink, try to create the new
73 * name indicated by the symlink. The old code always complained that the
74 * name already exists, due to not following the symlink even if its target
75 * is nonexistent. The new semantics affects also mknod() and link() when
76 * the name is a symlink pointing to a non-existent name.
78 * I don't know which semantics is the right one, since I have no access
79 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
81 * "old" one. Personally, I think the new semantics is much more logical.
82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
83 * file does succeed in both HP-UX and SunOs, but not in Solaris
84 * and in the old Linux semantics.
87 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
88 * semantics. See the comments in "open_namei" and "do_link" below.
90 * [10-Sep-98 Alan Modra] Another symlink change.
93 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
94 * inside the path - always follow.
95 * in the last component in creation/removal/renaming - never follow.
96 * if LOOKUP_FOLLOW passed - follow.
97 * if the pathname has trailing slashes - follow.
98 * otherwise - don't follow.
99 * (applied in that order).
101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
103 * During the 2.4 we need to fix the userland stuff depending on it -
104 * hopefully we will be able to get rid of that wart in 2.5. So far only
105 * XEmacs seems to be relying on it...
108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
109 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
110 * any extra contention...
113 /* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
120 static char *getname_flags(const char __user *filename, int flags, int *empty)
122 char *result = __getname(), *err;
125 if (unlikely(!result))
126 return ERR_PTR(-ENOMEM);
128 len = strncpy_from_user(result, filename, PATH_MAX);
130 if (unlikely(len < 0))
133 /* The empty path is special. */
134 if (unlikely(!len)) {
137 err = ERR_PTR(-ENOENT);
138 if (!(flags & LOOKUP_EMPTY))
142 err = ERR_PTR(-ENAMETOOLONG);
143 if (likely(len < PATH_MAX)) {
144 audit_getname(result);
153 char *getname(const char __user * filename)
155 return getname_flags(filename, 0, NULL);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name)
161 if (unlikely(!audit_dummy_context()))
166 EXPORT_SYMBOL(putname);
169 static int check_acl(struct inode *inode, int mask)
171 #ifdef CONFIG_FS_POSIX_ACL
172 struct posix_acl *acl;
174 if (mask & MAY_NOT_BLOCK) {
175 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
178 /* no ->get_acl() calls in RCU mode... */
179 if (acl == ACL_NOT_CACHED)
181 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
184 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
187 * A filesystem can force a ACL callback by just never filling the
188 * ACL cache. But normally you'd fill the cache either at inode
189 * instantiation time, or on the first ->get_acl call.
191 * If the filesystem doesn't have a get_acl() function at all, we'll
192 * just create the negative cache entry.
194 if (acl == ACL_NOT_CACHED) {
195 if (inode->i_op->get_acl) {
196 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
200 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
206 int error = posix_acl_permission(inode, acl, mask);
207 posix_acl_release(acl);
216 * This does the basic permission checking
218 static int acl_permission_check(struct inode *inode, int mask)
220 unsigned int mode = inode->i_mode;
222 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
225 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
226 int error = check_acl(inode, mask);
227 if (error != -EAGAIN)
231 if (in_group_p(inode->i_gid))
236 * If the DACs are ok we don't need any capability check.
238 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
244 * generic_permission - check for access rights on a Posix-like filesystem
245 * @inode: inode to check access rights for
246 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
248 * Used to check for read/write/execute permissions on a file.
249 * We use "fsuid" for this, letting us set arbitrary permissions
250 * for filesystem access without changing the "normal" uids which
251 * are used for other things.
253 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
254 * request cannot be satisfied (eg. requires blocking or too much complexity).
255 * It would then be called again in ref-walk mode.
257 int generic_permission(struct inode *inode, int mask)
262 * Do the basic permission checks.
264 ret = acl_permission_check(inode, mask);
268 if (S_ISDIR(inode->i_mode)) {
269 /* DACs are overridable for directories */
270 if (inode_capable(inode, CAP_DAC_OVERRIDE))
272 if (!(mask & MAY_WRITE))
273 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
278 * Read/write DACs are always overridable.
279 * Executable DACs are overridable when there is
280 * at least one exec bit set.
282 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
283 if (inode_capable(inode, CAP_DAC_OVERRIDE))
287 * Searching includes executable on directories, else just read.
289 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
290 if (mask == MAY_READ)
291 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
298 * We _really_ want to just do "generic_permission()" without
299 * even looking at the inode->i_op values. So we keep a cache
300 * flag in inode->i_opflags, that says "this has not special
301 * permission function, use the fast case".
303 static inline int do_inode_permission(struct inode *inode, int mask)
305 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
306 if (likely(inode->i_op->permission))
307 return inode->i_op->permission(inode, mask);
309 /* This gets set once for the inode lifetime */
310 spin_lock(&inode->i_lock);
311 inode->i_opflags |= IOP_FASTPERM;
312 spin_unlock(&inode->i_lock);
314 return generic_permission(inode, mask);
318 * __inode_permission - Check for access rights to a given inode
319 * @inode: Inode to check permission on
320 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
322 * Check for read/write/execute permissions on an inode.
324 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
326 * This does not check for a read-only file system. You probably want
327 * inode_permission().
329 int __inode_permission(struct inode *inode, int mask)
333 if (unlikely(mask & MAY_WRITE)) {
335 * Nobody gets write access to an immutable file.
337 if (IS_IMMUTABLE(inode))
341 retval = do_inode_permission(inode, mask);
345 retval = devcgroup_inode_permission(inode, mask);
349 return security_inode_permission(inode, mask);
353 * sb_permission - Check superblock-level permissions
354 * @sb: Superblock of inode to check permission on
355 * @inode: Inode to check permission on
356 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
358 * Separate out file-system wide checks from inode-specific permission checks.
360 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
362 if (unlikely(mask & MAY_WRITE)) {
363 umode_t mode = inode->i_mode;
365 /* Nobody gets write access to a read-only fs. */
366 if ((sb->s_flags & MS_RDONLY) &&
367 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
374 * inode_permission - Check for access rights to a given inode
375 * @inode: Inode to check permission on
376 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
378 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
379 * this, letting us set arbitrary permissions for filesystem access without
380 * changing the "normal" UIDs which are used for other things.
382 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
384 int inode_permission(struct inode *inode, int mask)
388 retval = sb_permission(inode->i_sb, inode, mask);
391 return __inode_permission(inode, mask);
395 * path_get - get a reference to a path
396 * @path: path to get the reference to
398 * Given a path increment the reference count to the dentry and the vfsmount.
400 void path_get(struct path *path)
405 EXPORT_SYMBOL(path_get);
408 * path_put - put a reference to a path
409 * @path: path to put the reference to
411 * Given a path decrement the reference count to the dentry and the vfsmount.
413 void path_put(struct path *path)
418 EXPORT_SYMBOL(path_put);
421 * Path walking has 2 modes, rcu-walk and ref-walk (see
422 * Documentation/filesystems/path-lookup.txt). In situations when we can't
423 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
424 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
425 * mode. Refcounts are grabbed at the last known good point before rcu-walk
426 * got stuck, so ref-walk may continue from there. If this is not successful
427 * (eg. a seqcount has changed), then failure is returned and it's up to caller
428 * to restart the path walk from the beginning in ref-walk mode.
431 static inline void lock_rcu_walk(void)
433 br_read_lock(&vfsmount_lock);
437 static inline void unlock_rcu_walk(void)
440 br_read_unlock(&vfsmount_lock);
444 * unlazy_walk - try to switch to ref-walk mode.
445 * @nd: nameidata pathwalk data
446 * @dentry: child of nd->path.dentry or NULL
447 * Returns: 0 on success, -ECHILD on failure
449 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
450 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
451 * @nd or NULL. Must be called from rcu-walk context.
453 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
455 struct fs_struct *fs = current->fs;
456 struct dentry *parent = nd->path.dentry;
459 BUG_ON(!(nd->flags & LOOKUP_RCU));
460 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
462 spin_lock(&fs->lock);
463 if (nd->root.mnt != fs->root.mnt ||
464 nd->root.dentry != fs->root.dentry)
467 spin_lock(&parent->d_lock);
469 if (!__d_rcu_to_refcount(parent, nd->seq))
471 BUG_ON(nd->inode != parent->d_inode);
473 if (dentry->d_parent != parent)
475 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
476 if (!__d_rcu_to_refcount(dentry, nd->seq))
479 * If the sequence check on the child dentry passed, then
480 * the child has not been removed from its parent. This
481 * means the parent dentry must be valid and able to take
482 * a reference at this point.
484 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
485 BUG_ON(!parent->d_count);
487 spin_unlock(&dentry->d_lock);
489 spin_unlock(&parent->d_lock);
492 spin_unlock(&fs->lock);
494 mntget(nd->path.mnt);
497 nd->flags &= ~LOOKUP_RCU;
501 spin_unlock(&dentry->d_lock);
503 spin_unlock(&parent->d_lock);
506 spin_unlock(&fs->lock);
510 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
512 return dentry->d_op->d_revalidate(dentry, flags);
516 * complete_walk - successful completion of path walk
517 * @nd: pointer nameidata
519 * If we had been in RCU mode, drop out of it and legitimize nd->path.
520 * Revalidate the final result, unless we'd already done that during
521 * the path walk or the filesystem doesn't ask for it. Return 0 on
522 * success, -error on failure. In case of failure caller does not
523 * need to drop nd->path.
525 static int complete_walk(struct nameidata *nd)
527 struct dentry *dentry = nd->path.dentry;
530 if (nd->flags & LOOKUP_RCU) {
531 nd->flags &= ~LOOKUP_RCU;
532 if (!(nd->flags & LOOKUP_ROOT))
534 spin_lock(&dentry->d_lock);
535 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
536 spin_unlock(&dentry->d_lock);
540 BUG_ON(nd->inode != dentry->d_inode);
541 spin_unlock(&dentry->d_lock);
542 mntget(nd->path.mnt);
546 if (likely(!(nd->flags & LOOKUP_JUMPED)))
549 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
552 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
555 /* Note: we do not d_invalidate() */
556 status = d_revalidate(dentry, nd->flags);
567 static __always_inline void set_root(struct nameidata *nd)
570 get_fs_root(current->fs, &nd->root);
573 static int link_path_walk(const char *, struct nameidata *);
575 static __always_inline void set_root_rcu(struct nameidata *nd)
578 struct fs_struct *fs = current->fs;
582 seq = read_seqcount_begin(&fs->seq);
584 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
585 } while (read_seqcount_retry(&fs->seq, seq));
589 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
601 nd->flags |= LOOKUP_JUMPED;
603 nd->inode = nd->path.dentry->d_inode;
605 ret = link_path_walk(link, nd);
609 return PTR_ERR(link);
612 static void path_put_conditional(struct path *path, struct nameidata *nd)
615 if (path->mnt != nd->path.mnt)
619 static inline void path_to_nameidata(const struct path *path,
620 struct nameidata *nd)
622 if (!(nd->flags & LOOKUP_RCU)) {
623 dput(nd->path.dentry);
624 if (nd->path.mnt != path->mnt)
625 mntput(nd->path.mnt);
627 nd->path.mnt = path->mnt;
628 nd->path.dentry = path->dentry;
632 * Helper to directly jump to a known parsed path from ->follow_link,
633 * caller must have taken a reference to path beforehand.
635 void nd_jump_link(struct nameidata *nd, struct path *path)
640 nd->inode = nd->path.dentry->d_inode;
641 nd->flags |= LOOKUP_JUMPED;
643 BUG_ON(nd->inode->i_op->follow_link);
646 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
648 struct inode *inode = link->dentry->d_inode;
649 if (inode->i_op->put_link)
650 inode->i_op->put_link(link->dentry, nd, cookie);
654 int sysctl_protected_symlinks __read_mostly = 1;
655 int sysctl_protected_hardlinks __read_mostly = 1;
658 * may_follow_link - Check symlink following for unsafe situations
659 * @link: The path of the symlink
660 * @nd: nameidata pathwalk data
662 * In the case of the sysctl_protected_symlinks sysctl being enabled,
663 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
664 * in a sticky world-writable directory. This is to protect privileged
665 * processes from failing races against path names that may change out
666 * from under them by way of other users creating malicious symlinks.
667 * It will permit symlinks to be followed only when outside a sticky
668 * world-writable directory, or when the uid of the symlink and follower
669 * match, or when the directory owner matches the symlink's owner.
671 * Returns 0 if following the symlink is allowed, -ve on error.
673 static inline int may_follow_link(struct path *link, struct nameidata *nd)
675 const struct inode *inode;
676 const struct inode *parent;
678 if (!sysctl_protected_symlinks)
681 /* Allowed if owner and follower match. */
682 inode = link->dentry->d_inode;
683 if (current_cred()->fsuid == inode->i_uid)
686 /* Allowed if parent directory not sticky and world-writable. */
687 parent = nd->path.dentry->d_inode;
688 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
691 /* Allowed if parent directory and link owner match. */
692 if (parent->i_uid == inode->i_uid)
695 path_put_conditional(link, nd);
697 audit_log_link_denied("follow_link", link);
702 * safe_hardlink_source - Check for safe hardlink conditions
703 * @inode: the source inode to hardlink from
705 * Return false if at least one of the following conditions:
706 * - inode is not a regular file
708 * - inode is setgid and group-exec
709 * - access failure for read and write
711 * Otherwise returns true.
713 static bool safe_hardlink_source(struct inode *inode)
715 umode_t mode = inode->i_mode;
717 /* Special files should not get pinned to the filesystem. */
721 /* Setuid files should not get pinned to the filesystem. */
725 /* Executable setgid files should not get pinned to the filesystem. */
726 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
729 /* Hardlinking to unreadable or unwritable sources is dangerous. */
730 if (inode_permission(inode, MAY_READ | MAY_WRITE))
737 * may_linkat - Check permissions for creating a hardlink
738 * @link: the source to hardlink from
740 * Block hardlink when all of:
741 * - sysctl_protected_hardlinks enabled
742 * - fsuid does not match inode
743 * - hardlink source is unsafe (see safe_hardlink_source() above)
746 * Returns 0 if successful, -ve on error.
748 static int may_linkat(struct path *link)
750 const struct cred *cred;
753 if (!sysctl_protected_hardlinks)
756 cred = current_cred();
757 inode = link->dentry->d_inode;
759 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
760 * otherwise, it must be a safe source.
762 if (cred->fsuid == inode->i_uid || safe_hardlink_source(inode) ||
766 audit_log_link_denied("linkat", link);
770 static __always_inline int
771 follow_link(struct path *link, struct nameidata *nd, void **p)
773 struct dentry *dentry = link->dentry;
777 BUG_ON(nd->flags & LOOKUP_RCU);
779 if (link->mnt == nd->path.mnt)
783 if (unlikely(current->total_link_count >= 40))
784 goto out_put_nd_path;
787 current->total_link_count++;
790 nd_set_link(nd, NULL);
792 error = security_inode_follow_link(link->dentry, nd);
794 goto out_put_nd_path;
796 nd->last_type = LAST_BIND;
797 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
800 goto out_put_nd_path;
805 error = __vfs_follow_link(nd, s);
807 put_link(nd, link, *p);
818 static int follow_up_rcu(struct path *path)
820 struct mount *mnt = real_mount(path->mnt);
821 struct mount *parent;
822 struct dentry *mountpoint;
824 parent = mnt->mnt_parent;
825 if (&parent->mnt == path->mnt)
827 mountpoint = mnt->mnt_mountpoint;
828 path->dentry = mountpoint;
829 path->mnt = &parent->mnt;
834 * follow_up - Find the mountpoint of path's vfsmount
836 * Given a path, find the mountpoint of its source file system.
837 * Replace @path with the path of the mountpoint in the parent mount.
840 * Return 1 if we went up a level and 0 if we were already at the
843 int follow_up(struct path *path)
845 struct mount *mnt = real_mount(path->mnt);
846 struct mount *parent;
847 struct dentry *mountpoint;
849 br_read_lock(&vfsmount_lock);
850 parent = mnt->mnt_parent;
852 br_read_unlock(&vfsmount_lock);
855 mntget(&parent->mnt);
856 mountpoint = dget(mnt->mnt_mountpoint);
857 br_read_unlock(&vfsmount_lock);
859 path->dentry = mountpoint;
861 path->mnt = &parent->mnt;
866 * Perform an automount
867 * - return -EISDIR to tell follow_managed() to stop and return the path we
870 static int follow_automount(struct path *path, unsigned flags,
873 struct vfsmount *mnt;
876 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
879 /* We don't want to mount if someone's just doing a stat -
880 * unless they're stat'ing a directory and appended a '/' to
883 * We do, however, want to mount if someone wants to open or
884 * create a file of any type under the mountpoint, wants to
885 * traverse through the mountpoint or wants to open the
886 * mounted directory. Also, autofs may mark negative dentries
887 * as being automount points. These will need the attentions
888 * of the daemon to instantiate them before they can be used.
890 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
891 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
892 path->dentry->d_inode)
895 current->total_link_count++;
896 if (current->total_link_count >= 40)
899 mnt = path->dentry->d_op->d_automount(path);
902 * The filesystem is allowed to return -EISDIR here to indicate
903 * it doesn't want to automount. For instance, autofs would do
904 * this so that its userspace daemon can mount on this dentry.
906 * However, we can only permit this if it's a terminal point in
907 * the path being looked up; if it wasn't then the remainder of
908 * the path is inaccessible and we should say so.
910 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
915 if (!mnt) /* mount collision */
919 /* lock_mount() may release path->mnt on error */
923 err = finish_automount(mnt, path);
927 /* Someone else made a mount here whilst we were busy */
932 path->dentry = dget(mnt->mnt_root);
941 * Handle a dentry that is managed in some way.
942 * - Flagged for transit management (autofs)
943 * - Flagged as mountpoint
944 * - Flagged as automount point
946 * This may only be called in refwalk mode.
948 * Serialization is taken care of in namespace.c
950 static int follow_managed(struct path *path, unsigned flags)
952 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
954 bool need_mntput = false;
957 /* Given that we're not holding a lock here, we retain the value in a
958 * local variable for each dentry as we look at it so that we don't see
959 * the components of that value change under us */
960 while (managed = ACCESS_ONCE(path->dentry->d_flags),
961 managed &= DCACHE_MANAGED_DENTRY,
962 unlikely(managed != 0)) {
963 /* Allow the filesystem to manage the transit without i_mutex
965 if (managed & DCACHE_MANAGE_TRANSIT) {
966 BUG_ON(!path->dentry->d_op);
967 BUG_ON(!path->dentry->d_op->d_manage);
968 ret = path->dentry->d_op->d_manage(path->dentry, false);
973 /* Transit to a mounted filesystem. */
974 if (managed & DCACHE_MOUNTED) {
975 struct vfsmount *mounted = lookup_mnt(path);
981 path->dentry = dget(mounted->mnt_root);
986 /* Something is mounted on this dentry in another
987 * namespace and/or whatever was mounted there in this
988 * namespace got unmounted before we managed to get the
992 /* Handle an automount point */
993 if (managed & DCACHE_NEED_AUTOMOUNT) {
994 ret = follow_automount(path, flags, &need_mntput);
1000 /* We didn't change the current path point */
1004 if (need_mntput && path->mnt == mnt)
1008 return ret < 0 ? ret : need_mntput;
1011 int follow_down_one(struct path *path)
1013 struct vfsmount *mounted;
1015 mounted = lookup_mnt(path);
1019 path->mnt = mounted;
1020 path->dentry = dget(mounted->mnt_root);
1026 static inline bool managed_dentry_might_block(struct dentry *dentry)
1028 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
1029 dentry->d_op->d_manage(dentry, true) < 0);
1033 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1034 * we meet a managed dentry that would need blocking.
1036 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1037 struct inode **inode)
1040 struct mount *mounted;
1042 * Don't forget we might have a non-mountpoint managed dentry
1043 * that wants to block transit.
1045 if (unlikely(managed_dentry_might_block(path->dentry)))
1048 if (!d_mountpoint(path->dentry))
1051 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1054 path->mnt = &mounted->mnt;
1055 path->dentry = mounted->mnt.mnt_root;
1056 nd->flags |= LOOKUP_JUMPED;
1057 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1059 * Update the inode too. We don't need to re-check the
1060 * dentry sequence number here after this d_inode read,
1061 * because a mount-point is always pinned.
1063 *inode = path->dentry->d_inode;
1068 static void follow_mount_rcu(struct nameidata *nd)
1070 while (d_mountpoint(nd->path.dentry)) {
1071 struct mount *mounted;
1072 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
1075 nd->path.mnt = &mounted->mnt;
1076 nd->path.dentry = mounted->mnt.mnt_root;
1077 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1081 static int follow_dotdot_rcu(struct nameidata *nd)
1086 if (nd->path.dentry == nd->root.dentry &&
1087 nd->path.mnt == nd->root.mnt) {
1090 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1091 struct dentry *old = nd->path.dentry;
1092 struct dentry *parent = old->d_parent;
1095 seq = read_seqcount_begin(&parent->d_seq);
1096 if (read_seqcount_retry(&old->d_seq, nd->seq))
1098 nd->path.dentry = parent;
1102 if (!follow_up_rcu(&nd->path))
1104 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1106 follow_mount_rcu(nd);
1107 nd->inode = nd->path.dentry->d_inode;
1111 nd->flags &= ~LOOKUP_RCU;
1112 if (!(nd->flags & LOOKUP_ROOT))
1113 nd->root.mnt = NULL;
1119 * Follow down to the covering mount currently visible to userspace. At each
1120 * point, the filesystem owning that dentry may be queried as to whether the
1121 * caller is permitted to proceed or not.
1123 int follow_down(struct path *path)
1128 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1129 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1130 /* Allow the filesystem to manage the transit without i_mutex
1133 * We indicate to the filesystem if someone is trying to mount
1134 * something here. This gives autofs the chance to deny anyone
1135 * other than its daemon the right to mount on its
1138 * The filesystem may sleep at this point.
1140 if (managed & DCACHE_MANAGE_TRANSIT) {
1141 BUG_ON(!path->dentry->d_op);
1142 BUG_ON(!path->dentry->d_op->d_manage);
1143 ret = path->dentry->d_op->d_manage(
1144 path->dentry, false);
1146 return ret == -EISDIR ? 0 : ret;
1149 /* Transit to a mounted filesystem. */
1150 if (managed & DCACHE_MOUNTED) {
1151 struct vfsmount *mounted = lookup_mnt(path);
1156 path->mnt = mounted;
1157 path->dentry = dget(mounted->mnt_root);
1161 /* Don't handle automount points here */
1168 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1170 static void follow_mount(struct path *path)
1172 while (d_mountpoint(path->dentry)) {
1173 struct vfsmount *mounted = lookup_mnt(path);
1178 path->mnt = mounted;
1179 path->dentry = dget(mounted->mnt_root);
1183 static void follow_dotdot(struct nameidata *nd)
1188 struct dentry *old = nd->path.dentry;
1190 if (nd->path.dentry == nd->root.dentry &&
1191 nd->path.mnt == nd->root.mnt) {
1194 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1195 /* rare case of legitimate dget_parent()... */
1196 nd->path.dentry = dget_parent(nd->path.dentry);
1200 if (!follow_up(&nd->path))
1203 follow_mount(&nd->path);
1204 nd->inode = nd->path.dentry->d_inode;
1208 * This looks up the name in dcache, possibly revalidates the old dentry and
1209 * allocates a new one if not found or not valid. In the need_lookup argument
1210 * returns whether i_op->lookup is necessary.
1212 * dir->d_inode->i_mutex must be held
1214 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1215 unsigned int flags, bool *need_lookup)
1217 struct dentry *dentry;
1220 *need_lookup = false;
1221 dentry = d_lookup(dir, name);
1223 if (d_need_lookup(dentry)) {
1224 *need_lookup = true;
1225 } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1226 error = d_revalidate(dentry, flags);
1227 if (unlikely(error <= 0)) {
1230 return ERR_PTR(error);
1231 } else if (!d_invalidate(dentry)) {
1240 dentry = d_alloc(dir, name);
1241 if (unlikely(!dentry))
1242 return ERR_PTR(-ENOMEM);
1244 *need_lookup = true;
1250 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1251 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1253 * dir->d_inode->i_mutex must be held
1255 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1260 /* Don't create child dentry for a dead directory. */
1261 if (unlikely(IS_DEADDIR(dir))) {
1263 return ERR_PTR(-ENOENT);
1266 old = dir->i_op->lookup(dir, dentry, flags);
1267 if (unlikely(old)) {
1274 static struct dentry *__lookup_hash(struct qstr *name,
1275 struct dentry *base, unsigned int flags)
1278 struct dentry *dentry;
1280 dentry = lookup_dcache(name, base, flags, &need_lookup);
1284 return lookup_real(base->d_inode, dentry, flags);
1288 * It's more convoluted than I'd like it to be, but... it's still fairly
1289 * small and for now I'd prefer to have fast path as straight as possible.
1290 * It _is_ time-critical.
1292 static int lookup_fast(struct nameidata *nd, struct qstr *name,
1293 struct path *path, struct inode **inode)
1295 struct vfsmount *mnt = nd->path.mnt;
1296 struct dentry *dentry, *parent = nd->path.dentry;
1302 * Rename seqlock is not required here because in the off chance
1303 * of a false negative due to a concurrent rename, we're going to
1304 * do the non-racy lookup, below.
1306 if (nd->flags & LOOKUP_RCU) {
1308 dentry = __d_lookup_rcu(parent, name, &seq, nd->inode);
1313 * This sequence count validates that the inode matches
1314 * the dentry name information from lookup.
1316 *inode = dentry->d_inode;
1317 if (read_seqcount_retry(&dentry->d_seq, seq))
1321 * This sequence count validates that the parent had no
1322 * changes while we did the lookup of the dentry above.
1324 * The memory barrier in read_seqcount_begin of child is
1325 * enough, we can use __read_seqcount_retry here.
1327 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1331 if (unlikely(d_need_lookup(dentry)))
1333 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1334 status = d_revalidate(dentry, nd->flags);
1335 if (unlikely(status <= 0)) {
1336 if (status != -ECHILD)
1342 path->dentry = dentry;
1343 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1345 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1349 if (unlazy_walk(nd, dentry))
1352 dentry = __d_lookup(parent, name);
1355 if (unlikely(!dentry))
1358 if (unlikely(d_need_lookup(dentry))) {
1363 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1364 status = d_revalidate(dentry, nd->flags);
1365 if (unlikely(status <= 0)) {
1370 if (!d_invalidate(dentry)) {
1377 path->dentry = dentry;
1378 err = follow_managed(path, nd->flags);
1379 if (unlikely(err < 0)) {
1380 path_put_conditional(path, nd);
1384 nd->flags |= LOOKUP_JUMPED;
1385 *inode = path->dentry->d_inode;
1392 /* Fast lookup failed, do it the slow way */
1393 static int lookup_slow(struct nameidata *nd, struct qstr *name,
1396 struct dentry *dentry, *parent;
1399 parent = nd->path.dentry;
1400 BUG_ON(nd->inode != parent->d_inode);
1402 mutex_lock(&parent->d_inode->i_mutex);
1403 dentry = __lookup_hash(name, parent, nd->flags);
1404 mutex_unlock(&parent->d_inode->i_mutex);
1406 return PTR_ERR(dentry);
1407 path->mnt = nd->path.mnt;
1408 path->dentry = dentry;
1409 err = follow_managed(path, nd->flags);
1410 if (unlikely(err < 0)) {
1411 path_put_conditional(path, nd);
1415 nd->flags |= LOOKUP_JUMPED;
1419 static inline int may_lookup(struct nameidata *nd)
1421 if (nd->flags & LOOKUP_RCU) {
1422 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1425 if (unlazy_walk(nd, NULL))
1428 return inode_permission(nd->inode, MAY_EXEC);
1431 static inline int handle_dots(struct nameidata *nd, int type)
1433 if (type == LAST_DOTDOT) {
1434 if (nd->flags & LOOKUP_RCU) {
1435 if (follow_dotdot_rcu(nd))
1443 static void terminate_walk(struct nameidata *nd)
1445 if (!(nd->flags & LOOKUP_RCU)) {
1446 path_put(&nd->path);
1448 nd->flags &= ~LOOKUP_RCU;
1449 if (!(nd->flags & LOOKUP_ROOT))
1450 nd->root.mnt = NULL;
1456 * Do we need to follow links? We _really_ want to be able
1457 * to do this check without having to look at inode->i_op,
1458 * so we keep a cache of "no, this doesn't need follow_link"
1459 * for the common case.
1461 static inline int should_follow_link(struct inode *inode, int follow)
1463 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1464 if (likely(inode->i_op->follow_link))
1467 /* This gets set once for the inode lifetime */
1468 spin_lock(&inode->i_lock);
1469 inode->i_opflags |= IOP_NOFOLLOW;
1470 spin_unlock(&inode->i_lock);
1475 static inline int walk_component(struct nameidata *nd, struct path *path,
1476 struct qstr *name, int type, int follow)
1478 struct inode *inode;
1481 * "." and ".." are special - ".." especially so because it has
1482 * to be able to know about the current root directory and
1483 * parent relationships.
1485 if (unlikely(type != LAST_NORM))
1486 return handle_dots(nd, type);
1487 err = lookup_fast(nd, name, path, &inode);
1488 if (unlikely(err)) {
1492 err = lookup_slow(nd, name, path);
1496 inode = path->dentry->d_inode;
1502 if (should_follow_link(inode, follow)) {
1503 if (nd->flags & LOOKUP_RCU) {
1504 if (unlikely(unlazy_walk(nd, path->dentry))) {
1509 BUG_ON(inode != path->dentry->d_inode);
1512 path_to_nameidata(path, nd);
1517 path_to_nameidata(path, nd);
1524 * This limits recursive symlink follows to 8, while
1525 * limiting consecutive symlinks to 40.
1527 * Without that kind of total limit, nasty chains of consecutive
1528 * symlinks can cause almost arbitrarily long lookups.
1530 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1534 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1535 path_put_conditional(path, nd);
1536 path_put(&nd->path);
1539 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1542 current->link_count++;
1545 struct path link = *path;
1548 res = follow_link(&link, nd, &cookie);
1551 res = walk_component(nd, path, &nd->last,
1552 nd->last_type, LOOKUP_FOLLOW);
1553 put_link(nd, &link, cookie);
1556 current->link_count--;
1562 * We really don't want to look at inode->i_op->lookup
1563 * when we don't have to. So we keep a cache bit in
1564 * the inode ->i_opflags field that says "yes, we can
1565 * do lookup on this inode".
1567 static inline int can_lookup(struct inode *inode)
1569 if (likely(inode->i_opflags & IOP_LOOKUP))
1571 if (likely(!inode->i_op->lookup))
1574 /* We do this once for the lifetime of the inode */
1575 spin_lock(&inode->i_lock);
1576 inode->i_opflags |= IOP_LOOKUP;
1577 spin_unlock(&inode->i_lock);
1582 * We can do the critical dentry name comparison and hashing
1583 * operations one word at a time, but we are limited to:
1585 * - Architectures with fast unaligned word accesses. We could
1586 * do a "get_unaligned()" if this helps and is sufficiently
1589 * - Little-endian machines (so that we can generate the mask
1590 * of low bytes efficiently). Again, we *could* do a byte
1591 * swapping load on big-endian architectures if that is not
1592 * expensive enough to make the optimization worthless.
1594 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1595 * do not trap on the (extremely unlikely) case of a page
1596 * crossing operation.
1598 * - Furthermore, we need an efficient 64-bit compile for the
1599 * 64-bit case in order to generate the "number of bytes in
1600 * the final mask". Again, that could be replaced with a
1601 * efficient population count instruction or similar.
1603 #ifdef CONFIG_DCACHE_WORD_ACCESS
1605 #include <asm/word-at-a-time.h>
1609 static inline unsigned int fold_hash(unsigned long hash)
1611 hash += hash >> (8*sizeof(int));
1615 #else /* 32-bit case */
1617 #define fold_hash(x) (x)
1621 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1623 unsigned long a, mask;
1624 unsigned long hash = 0;
1627 a = load_unaligned_zeropad(name);
1628 if (len < sizeof(unsigned long))
1632 name += sizeof(unsigned long);
1633 len -= sizeof(unsigned long);
1637 mask = ~(~0ul << len*8);
1640 return fold_hash(hash);
1642 EXPORT_SYMBOL(full_name_hash);
1645 * Calculate the length and hash of the path component, and
1646 * return the length of the component;
1648 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1650 unsigned long a, b, adata, bdata, mask, hash, len;
1651 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1654 len = -sizeof(unsigned long);
1656 hash = (hash + a) * 9;
1657 len += sizeof(unsigned long);
1658 a = load_unaligned_zeropad(name+len);
1659 b = a ^ REPEAT_BYTE('/');
1660 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1662 adata = prep_zero_mask(a, adata, &constants);
1663 bdata = prep_zero_mask(b, bdata, &constants);
1665 mask = create_zero_mask(adata | bdata);
1667 hash += a & zero_bytemask(mask);
1668 *hashp = fold_hash(hash);
1670 return len + find_zero(mask);
1675 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1677 unsigned long hash = init_name_hash();
1679 hash = partial_name_hash(*name++, hash);
1680 return end_name_hash(hash);
1682 EXPORT_SYMBOL(full_name_hash);
1685 * We know there's a real path component here of at least
1688 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1690 unsigned long hash = init_name_hash();
1691 unsigned long len = 0, c;
1693 c = (unsigned char)*name;
1696 hash = partial_name_hash(c, hash);
1697 c = (unsigned char)name[len];
1698 } while (c && c != '/');
1699 *hashp = end_name_hash(hash);
1707 * This is the basic name resolution function, turning a pathname into
1708 * the final dentry. We expect 'base' to be positive and a directory.
1710 * Returns 0 and nd will have valid dentry and mnt on success.
1711 * Returns error and drops reference to input namei data on failure.
1713 static int link_path_walk(const char *name, struct nameidata *nd)
1723 /* At this point we know we have a real path component. */
1729 err = may_lookup(nd);
1733 len = hash_name(name, &this.hash);
1738 if (name[0] == '.') switch (len) {
1740 if (name[1] == '.') {
1742 nd->flags |= LOOKUP_JUMPED;
1748 if (likely(type == LAST_NORM)) {
1749 struct dentry *parent = nd->path.dentry;
1750 nd->flags &= ~LOOKUP_JUMPED;
1751 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1752 err = parent->d_op->d_hash(parent, nd->inode,
1760 goto last_component;
1762 * If it wasn't NUL, we know it was '/'. Skip that
1763 * slash, and continue until no more slashes.
1767 } while (unlikely(name[len] == '/'));
1769 goto last_component;
1772 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1777 err = nested_symlink(&next, nd);
1781 if (can_lookup(nd->inode))
1785 /* here ends the main loop */
1789 nd->last_type = type;
1796 static int path_init(int dfd, const char *name, unsigned int flags,
1797 struct nameidata *nd, struct file **fp)
1803 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1804 nd->flags = flags | LOOKUP_JUMPED;
1806 if (flags & LOOKUP_ROOT) {
1807 struct inode *inode = nd->root.dentry->d_inode;
1809 if (!inode->i_op->lookup)
1811 retval = inode_permission(inode, MAY_EXEC);
1815 nd->path = nd->root;
1817 if (flags & LOOKUP_RCU) {
1819 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1821 path_get(&nd->path);
1826 nd->root.mnt = NULL;
1829 if (flags & LOOKUP_RCU) {
1834 path_get(&nd->root);
1836 nd->path = nd->root;
1837 } else if (dfd == AT_FDCWD) {
1838 if (flags & LOOKUP_RCU) {
1839 struct fs_struct *fs = current->fs;
1845 seq = read_seqcount_begin(&fs->seq);
1847 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1848 } while (read_seqcount_retry(&fs->seq, seq));
1850 get_fs_pwd(current->fs, &nd->path);
1853 struct dentry *dentry;
1855 file = fget_raw_light(dfd, &fput_needed);
1860 dentry = file->f_path.dentry;
1864 if (!S_ISDIR(dentry->d_inode->i_mode))
1867 retval = inode_permission(dentry->d_inode, MAY_EXEC);
1872 nd->path = file->f_path;
1873 if (flags & LOOKUP_RCU) {
1876 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1879 path_get(&file->f_path);
1880 fput_light(file, fput_needed);
1884 nd->inode = nd->path.dentry->d_inode;
1888 fput_light(file, fput_needed);
1893 static inline int lookup_last(struct nameidata *nd, struct path *path)
1895 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1896 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1898 nd->flags &= ~LOOKUP_PARENT;
1899 return walk_component(nd, path, &nd->last, nd->last_type,
1900 nd->flags & LOOKUP_FOLLOW);
1903 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1904 static int path_lookupat(int dfd, const char *name,
1905 unsigned int flags, struct nameidata *nd)
1907 struct file *base = NULL;
1912 * Path walking is largely split up into 2 different synchronisation
1913 * schemes, rcu-walk and ref-walk (explained in
1914 * Documentation/filesystems/path-lookup.txt). These share much of the
1915 * path walk code, but some things particularly setup, cleanup, and
1916 * following mounts are sufficiently divergent that functions are
1917 * duplicated. Typically there is a function foo(), and its RCU
1918 * analogue, foo_rcu().
1920 * -ECHILD is the error number of choice (just to avoid clashes) that
1921 * is returned if some aspect of an rcu-walk fails. Such an error must
1922 * be handled by restarting a traditional ref-walk (which will always
1923 * be able to complete).
1925 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1930 current->total_link_count = 0;
1931 err = link_path_walk(name, nd);
1933 if (!err && !(flags & LOOKUP_PARENT)) {
1934 err = lookup_last(nd, &path);
1937 struct path link = path;
1938 err = may_follow_link(&link, nd);
1941 nd->flags |= LOOKUP_PARENT;
1942 err = follow_link(&link, nd, &cookie);
1945 err = lookup_last(nd, &path);
1946 put_link(nd, &link, cookie);
1951 err = complete_walk(nd);
1953 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1954 if (!nd->inode->i_op->lookup) {
1955 path_put(&nd->path);
1963 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1964 path_put(&nd->root);
1965 nd->root.mnt = NULL;
1970 static int do_path_lookup(int dfd, const char *name,
1971 unsigned int flags, struct nameidata *nd)
1973 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1974 if (unlikely(retval == -ECHILD))
1975 retval = path_lookupat(dfd, name, flags, nd);
1976 if (unlikely(retval == -ESTALE))
1977 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1979 if (likely(!retval)) {
1980 if (unlikely(!audit_dummy_context())) {
1981 if (nd->path.dentry && nd->inode)
1982 audit_inode(name, nd->path.dentry);
1988 /* does lookup, returns the object with parent locked */
1989 struct dentry *kern_path_locked(const char *name, struct path *path)
1991 struct nameidata nd;
1993 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
1995 return ERR_PTR(err);
1996 if (nd.last_type != LAST_NORM) {
1998 return ERR_PTR(-EINVAL);
2000 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2001 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2003 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2011 int kern_path(const char *name, unsigned int flags, struct path *path)
2013 struct nameidata nd;
2014 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2021 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2022 * @dentry: pointer to dentry of the base directory
2023 * @mnt: pointer to vfs mount of the base directory
2024 * @name: pointer to file name
2025 * @flags: lookup flags
2026 * @path: pointer to struct path to fill
2028 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2029 const char *name, unsigned int flags,
2032 struct nameidata nd;
2034 nd.root.dentry = dentry;
2036 BUG_ON(flags & LOOKUP_PARENT);
2037 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2038 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2045 * Restricted form of lookup. Doesn't follow links, single-component only,
2046 * needs parent already locked. Doesn't follow mounts.
2049 static struct dentry *lookup_hash(struct nameidata *nd)
2051 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2055 * lookup_one_len - filesystem helper to lookup single pathname component
2056 * @name: pathname component to lookup
2057 * @base: base directory to lookup from
2058 * @len: maximum length @len should be interpreted to
2060 * Note that this routine is purely a helper for filesystem usage and should
2061 * not be called by generic code. Also note that by using this function the
2062 * nameidata argument is passed to the filesystem methods and a filesystem
2063 * using this helper needs to be prepared for that.
2065 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2071 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2075 this.hash = full_name_hash(name, len);
2077 return ERR_PTR(-EACCES);
2080 c = *(const unsigned char *)name++;
2081 if (c == '/' || c == '\0')
2082 return ERR_PTR(-EACCES);
2085 * See if the low-level filesystem might want
2086 * to use its own hash..
2088 if (base->d_flags & DCACHE_OP_HASH) {
2089 int err = base->d_op->d_hash(base, base->d_inode, &this);
2091 return ERR_PTR(err);
2094 err = inode_permission(base->d_inode, MAY_EXEC);
2096 return ERR_PTR(err);
2098 return __lookup_hash(&this, base, 0);
2101 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2102 struct path *path, int *empty)
2104 struct nameidata nd;
2105 char *tmp = getname_flags(name, flags, empty);
2106 int err = PTR_ERR(tmp);
2109 BUG_ON(flags & LOOKUP_PARENT);
2111 err = do_path_lookup(dfd, tmp, flags, &nd);
2119 int user_path_at(int dfd, const char __user *name, unsigned flags,
2122 return user_path_at_empty(dfd, name, flags, path, NULL);
2125 static int user_path_parent(int dfd, const char __user *path,
2126 struct nameidata *nd, char **name)
2128 char *s = getname(path);
2134 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
2144 * It's inline, so penalty for filesystems that don't use sticky bit is
2147 static inline int check_sticky(struct inode *dir, struct inode *inode)
2149 kuid_t fsuid = current_fsuid();
2151 if (!(dir->i_mode & S_ISVTX))
2153 if (uid_eq(inode->i_uid, fsuid))
2155 if (uid_eq(dir->i_uid, fsuid))
2157 return !inode_capable(inode, CAP_FOWNER);
2161 * Check whether we can remove a link victim from directory dir, check
2162 * whether the type of victim is right.
2163 * 1. We can't do it if dir is read-only (done in permission())
2164 * 2. We should have write and exec permissions on dir
2165 * 3. We can't remove anything from append-only dir
2166 * 4. We can't do anything with immutable dir (done in permission())
2167 * 5. If the sticky bit on dir is set we should either
2168 * a. be owner of dir, or
2169 * b. be owner of victim, or
2170 * c. have CAP_FOWNER capability
2171 * 6. If the victim is append-only or immutable we can't do antyhing with
2172 * links pointing to it.
2173 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2174 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2175 * 9. We can't remove a root or mountpoint.
2176 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2177 * nfs_async_unlink().
2179 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
2183 if (!victim->d_inode)
2186 BUG_ON(victim->d_parent->d_inode != dir);
2187 audit_inode_child(victim, dir);
2189 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2194 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
2195 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
2198 if (!S_ISDIR(victim->d_inode->i_mode))
2200 if (IS_ROOT(victim))
2202 } else if (S_ISDIR(victim->d_inode->i_mode))
2204 if (IS_DEADDIR(dir))
2206 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2211 /* Check whether we can create an object with dentry child in directory
2213 * 1. We can't do it if child already exists (open has special treatment for
2214 * this case, but since we are inlined it's OK)
2215 * 2. We can't do it if dir is read-only (done in permission())
2216 * 3. We should have write and exec permissions on dir
2217 * 4. We can't do it if dir is immutable (done in permission())
2219 static inline int may_create(struct inode *dir, struct dentry *child)
2223 if (IS_DEADDIR(dir))
2225 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2229 * p1 and p2 should be directories on the same fs.
2231 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2236 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2240 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2242 p = d_ancestor(p2, p1);
2244 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2245 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2249 p = d_ancestor(p1, p2);
2251 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2252 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2256 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2257 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2261 void unlock_rename(struct dentry *p1, struct dentry *p2)
2263 mutex_unlock(&p1->d_inode->i_mutex);
2265 mutex_unlock(&p2->d_inode->i_mutex);
2266 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2270 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2273 int error = may_create(dir, dentry);
2277 if (!dir->i_op->create)
2278 return -EACCES; /* shouldn't it be ENOSYS? */
2281 error = security_inode_create(dir, dentry, mode);
2284 error = dir->i_op->create(dir, dentry, mode, want_excl);
2286 fsnotify_create(dir, dentry);
2290 static int may_open(struct path *path, int acc_mode, int flag)
2292 struct dentry *dentry = path->dentry;
2293 struct inode *inode = dentry->d_inode;
2303 switch (inode->i_mode & S_IFMT) {
2307 if (acc_mode & MAY_WRITE)
2312 if (path->mnt->mnt_flags & MNT_NODEV)
2321 error = inode_permission(inode, acc_mode);
2326 * An append-only file must be opened in append mode for writing.
2328 if (IS_APPEND(inode)) {
2329 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2335 /* O_NOATIME can only be set by the owner or superuser */
2336 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2342 static int handle_truncate(struct file *filp)
2344 struct path *path = &filp->f_path;
2345 struct inode *inode = path->dentry->d_inode;
2346 int error = get_write_access(inode);
2350 * Refuse to truncate files with mandatory locks held on them.
2352 error = locks_verify_locked(inode);
2354 error = security_path_truncate(path);
2356 error = do_truncate(path->dentry, 0,
2357 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2360 put_write_access(inode);
2364 static inline int open_to_namei_flags(int flag)
2366 if ((flag & O_ACCMODE) == 3)
2371 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2373 int error = security_path_mknod(dir, dentry, mode, 0);
2377 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2381 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2385 * Attempt to atomically look up, create and open a file from a negative
2388 * Returns 0 if successful. The file will have been created and attached to
2389 * @file by the filesystem calling finish_open().
2391 * Returns 1 if the file was looked up only or didn't need creating. The
2392 * caller will need to perform the open themselves. @path will have been
2393 * updated to point to the new dentry. This may be negative.
2395 * Returns an error code otherwise.
2397 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2398 struct path *path, struct file *file,
2399 const struct open_flags *op,
2400 bool got_write, bool need_lookup,
2403 struct inode *dir = nd->path.dentry->d_inode;
2404 unsigned open_flag = open_to_namei_flags(op->open_flag);
2408 int create_error = 0;
2409 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2411 BUG_ON(dentry->d_inode);
2413 /* Don't create child dentry for a dead directory. */
2414 if (unlikely(IS_DEADDIR(dir))) {
2420 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2421 mode &= ~current_umask();
2423 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) {
2424 open_flag &= ~O_TRUNC;
2425 *opened |= FILE_CREATED;
2429 * Checking write permission is tricky, bacuse we don't know if we are
2430 * going to actually need it: O_CREAT opens should work as long as the
2431 * file exists. But checking existence breaks atomicity. The trick is
2432 * to check access and if not granted clear O_CREAT from the flags.
2434 * Another problem is returing the "right" error value (e.g. for an
2435 * O_EXCL open we want to return EEXIST not EROFS).
2437 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2438 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2439 if (!(open_flag & O_CREAT)) {
2441 * No O_CREATE -> atomicity not a requirement -> fall
2442 * back to lookup + open
2445 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2446 /* Fall back and fail with the right error */
2447 create_error = -EROFS;
2450 /* No side effects, safe to clear O_CREAT */
2451 create_error = -EROFS;
2452 open_flag &= ~O_CREAT;
2456 if (open_flag & O_CREAT) {
2457 error = may_o_create(&nd->path, dentry, mode);
2459 create_error = error;
2460 if (open_flag & O_EXCL)
2462 open_flag &= ~O_CREAT;
2466 if (nd->flags & LOOKUP_DIRECTORY)
2467 open_flag |= O_DIRECTORY;
2469 file->f_path.dentry = DENTRY_NOT_SET;
2470 file->f_path.mnt = nd->path.mnt;
2471 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2474 if (create_error && error == -ENOENT)
2475 error = create_error;
2479 acc_mode = op->acc_mode;
2480 if (*opened & FILE_CREATED) {
2481 fsnotify_create(dir, dentry);
2482 acc_mode = MAY_OPEN;
2485 if (error) { /* returned 1, that is */
2486 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2490 if (file->f_path.dentry) {
2492 dentry = file->f_path.dentry;
2494 if (create_error && dentry->d_inode == NULL) {
2495 error = create_error;
2502 * We didn't have the inode before the open, so check open permission
2505 error = may_open(&file->f_path, acc_mode, open_flag);
2515 dentry = lookup_real(dir, dentry, nd->flags);
2517 return PTR_ERR(dentry);
2520 int open_flag = op->open_flag;
2522 error = create_error;
2523 if ((open_flag & O_EXCL)) {
2524 if (!dentry->d_inode)
2526 } else if (!dentry->d_inode) {
2528 } else if ((open_flag & O_TRUNC) &&
2529 S_ISREG(dentry->d_inode->i_mode)) {
2532 /* will fail later, go on to get the right error */
2536 path->dentry = dentry;
2537 path->mnt = nd->path.mnt;
2542 * Look up and maybe create and open the last component.
2544 * Must be called with i_mutex held on parent.
2546 * Returns 0 if the file was successfully atomically created (if necessary) and
2547 * opened. In this case the file will be returned attached to @file.
2549 * Returns 1 if the file was not completely opened at this time, though lookups
2550 * and creations will have been performed and the dentry returned in @path will
2551 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2552 * specified then a negative dentry may be returned.
2554 * An error code is returned otherwise.
2556 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2557 * cleared otherwise prior to returning.
2559 static int lookup_open(struct nameidata *nd, struct path *path,
2561 const struct open_flags *op,
2562 bool got_write, int *opened)
2564 struct dentry *dir = nd->path.dentry;
2565 struct inode *dir_inode = dir->d_inode;
2566 struct dentry *dentry;
2570 *opened &= ~FILE_CREATED;
2571 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2573 return PTR_ERR(dentry);
2575 /* Cached positive dentry: will open in f_op->open */
2576 if (!need_lookup && dentry->d_inode)
2579 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2580 return atomic_open(nd, dentry, path, file, op, got_write,
2581 need_lookup, opened);
2585 BUG_ON(dentry->d_inode);
2587 dentry = lookup_real(dir_inode, dentry, nd->flags);
2589 return PTR_ERR(dentry);
2592 /* Negative dentry, just create the file */
2593 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2594 umode_t mode = op->mode;
2595 if (!IS_POSIXACL(dir->d_inode))
2596 mode &= ~current_umask();
2598 * This write is needed to ensure that a
2599 * rw->ro transition does not occur between
2600 * the time when the file is created and when
2601 * a permanent write count is taken through
2602 * the 'struct file' in finish_open().
2608 *opened |= FILE_CREATED;
2609 error = security_path_mknod(&nd->path, dentry, mode, 0);
2612 error = vfs_create(dir->d_inode, dentry, mode,
2613 nd->flags & LOOKUP_EXCL);
2618 path->dentry = dentry;
2619 path->mnt = nd->path.mnt;
2628 * Handle the last step of open()
2630 static int do_last(struct nameidata *nd, struct path *path,
2631 struct file *file, const struct open_flags *op,
2632 int *opened, const char *pathname)
2634 struct dentry *dir = nd->path.dentry;
2635 int open_flag = op->open_flag;
2636 bool will_truncate = (open_flag & O_TRUNC) != 0;
2637 bool got_write = false;
2638 int acc_mode = op->acc_mode;
2639 struct inode *inode;
2640 bool symlink_ok = false;
2641 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2642 bool retried = false;
2645 nd->flags &= ~LOOKUP_PARENT;
2646 nd->flags |= op->intent;
2648 switch (nd->last_type) {
2651 error = handle_dots(nd, nd->last_type);
2656 error = complete_walk(nd);
2659 audit_inode(pathname, nd->path.dentry);
2660 if (open_flag & O_CREAT) {
2666 error = complete_walk(nd);
2669 audit_inode(pathname, dir);
2673 if (!(open_flag & O_CREAT)) {
2674 if (nd->last.name[nd->last.len])
2675 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2676 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2678 /* we _can_ be in RCU mode here */
2679 error = lookup_fast(nd, &nd->last, path, &inode);
2686 BUG_ON(nd->inode != dir->d_inode);
2688 /* create side of things */
2690 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2691 * has been cleared when we got to the last component we are
2694 error = complete_walk(nd);
2698 audit_inode(pathname, dir);
2700 /* trailing slashes? */
2701 if (nd->last.name[nd->last.len])
2706 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2707 error = mnt_want_write(nd->path.mnt);
2711 * do _not_ fail yet - we might not need that or fail with
2712 * a different error; let lookup_open() decide; we'll be
2713 * dropping this one anyway.
2716 mutex_lock(&dir->d_inode->i_mutex);
2717 error = lookup_open(nd, path, file, op, got_write, opened);
2718 mutex_unlock(&dir->d_inode->i_mutex);
2724 if ((*opened & FILE_CREATED) ||
2725 !S_ISREG(file->f_path.dentry->d_inode->i_mode))
2726 will_truncate = false;
2728 audit_inode(pathname, file->f_path.dentry);
2732 if (*opened & FILE_CREATED) {
2733 /* Don't check for write permission, don't truncate */
2734 open_flag &= ~O_TRUNC;
2735 will_truncate = false;
2736 acc_mode = MAY_OPEN;
2737 path_to_nameidata(path, nd);
2738 goto finish_open_created;
2742 * create/update audit record if it already exists.
2744 if (path->dentry->d_inode)
2745 audit_inode(pathname, path->dentry);
2748 * If atomic_open() acquired write access it is dropped now due to
2749 * possible mount and symlink following (this might be optimized away if
2753 mnt_drop_write(nd->path.mnt);
2758 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
2761 error = follow_managed(path, nd->flags);
2766 nd->flags |= LOOKUP_JUMPED;
2768 BUG_ON(nd->flags & LOOKUP_RCU);
2769 inode = path->dentry->d_inode;
2771 /* we _can_ be in RCU mode here */
2774 path_to_nameidata(path, nd);
2778 if (should_follow_link(inode, !symlink_ok)) {
2779 if (nd->flags & LOOKUP_RCU) {
2780 if (unlikely(unlazy_walk(nd, path->dentry))) {
2785 BUG_ON(inode != path->dentry->d_inode);
2789 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2790 path_to_nameidata(path, nd);
2792 save_parent.dentry = nd->path.dentry;
2793 save_parent.mnt = mntget(path->mnt);
2794 nd->path.dentry = path->dentry;
2798 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2799 error = complete_walk(nd);
2801 path_put(&save_parent);
2805 if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
2808 if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup)
2810 audit_inode(pathname, nd->path.dentry);
2812 if (!S_ISREG(nd->inode->i_mode))
2813 will_truncate = false;
2815 if (will_truncate) {
2816 error = mnt_want_write(nd->path.mnt);
2821 finish_open_created:
2822 error = may_open(&nd->path, acc_mode, open_flag);
2825 file->f_path.mnt = nd->path.mnt;
2826 error = finish_open(file, nd->path.dentry, NULL, opened);
2828 if (error == -EOPENSTALE)
2833 error = open_check_o_direct(file);
2836 error = ima_file_check(file, op->acc_mode);
2840 if (will_truncate) {
2841 error = handle_truncate(file);
2847 mnt_drop_write(nd->path.mnt);
2848 path_put(&save_parent);
2853 path_put_conditional(path, nd);
2860 /* If no saved parent or already retried then can't retry */
2861 if (!save_parent.dentry || retried)
2864 BUG_ON(save_parent.dentry != dir);
2865 path_put(&nd->path);
2866 nd->path = save_parent;
2867 nd->inode = dir->d_inode;
2868 save_parent.mnt = NULL;
2869 save_parent.dentry = NULL;
2871 mnt_drop_write(nd->path.mnt);
2878 static struct file *path_openat(int dfd, const char *pathname,
2879 struct nameidata *nd, const struct open_flags *op, int flags)
2881 struct file *base = NULL;
2887 file = get_empty_filp();
2889 return ERR_PTR(-ENFILE);
2891 file->f_flags = op->open_flag;
2893 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2894 if (unlikely(error))
2897 current->total_link_count = 0;
2898 error = link_path_walk(pathname, nd);
2899 if (unlikely(error))
2902 error = do_last(nd, &path, file, op, &opened, pathname);
2903 while (unlikely(error > 0)) { /* trailing symlink */
2904 struct path link = path;
2906 if (!(nd->flags & LOOKUP_FOLLOW)) {
2907 path_put_conditional(&path, nd);
2908 path_put(&nd->path);
2912 error = may_follow_link(&link, nd);
2913 if (unlikely(error))
2915 nd->flags |= LOOKUP_PARENT;
2916 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2917 error = follow_link(&link, nd, &cookie);
2918 if (unlikely(error))
2920 error = do_last(nd, &path, file, op, &opened, pathname);
2921 put_link(nd, &link, cookie);
2924 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2925 path_put(&nd->root);
2928 if (!(opened & FILE_OPENED)) {
2932 if (unlikely(error)) {
2933 if (error == -EOPENSTALE) {
2934 if (flags & LOOKUP_RCU)
2939 file = ERR_PTR(error);
2944 struct file *do_filp_open(int dfd, const char *pathname,
2945 const struct open_flags *op, int flags)
2947 struct nameidata nd;
2950 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2951 if (unlikely(filp == ERR_PTR(-ECHILD)))
2952 filp = path_openat(dfd, pathname, &nd, op, flags);
2953 if (unlikely(filp == ERR_PTR(-ESTALE)))
2954 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2958 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2959 const char *name, const struct open_flags *op, int flags)
2961 struct nameidata nd;
2965 nd.root.dentry = dentry;
2967 flags |= LOOKUP_ROOT;
2969 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2970 return ERR_PTR(-ELOOP);
2972 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2973 if (unlikely(file == ERR_PTR(-ECHILD)))
2974 file = path_openat(-1, name, &nd, op, flags);
2975 if (unlikely(file == ERR_PTR(-ESTALE)))
2976 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2980 struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
2982 struct dentry *dentry = ERR_PTR(-EEXIST);
2983 struct nameidata nd;
2985 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2987 return ERR_PTR(error);
2990 * Yucky last component or no last component at all?
2991 * (foo/., foo/.., /////)
2993 if (nd.last_type != LAST_NORM)
2995 nd.flags &= ~LOOKUP_PARENT;
2996 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2998 /* don't fail immediately if it's r/o, at least try to report other errors */
2999 err2 = mnt_want_write(nd.path.mnt);
3001 * Do the final lookup.
3003 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3004 dentry = lookup_hash(&nd);
3009 if (dentry->d_inode)
3012 * Special case - lookup gave negative, but... we had foo/bar/
3013 * From the vfs_mknod() POV we just have a negative dentry -
3014 * all is fine. Let's be bastards - you had / on the end, you've
3015 * been asking for (non-existent) directory. -ENOENT for you.
3017 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3021 if (unlikely(err2)) {
3029 dentry = ERR_PTR(error);
3031 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3033 mnt_drop_write(nd.path.mnt);
3038 EXPORT_SYMBOL(kern_path_create);
3040 void done_path_create(struct path *path, struct dentry *dentry)
3043 mutex_unlock(&path->dentry->d_inode->i_mutex);
3044 mnt_drop_write(path->mnt);
3047 EXPORT_SYMBOL(done_path_create);
3049 struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
3051 char *tmp = getname(pathname);
3054 return ERR_CAST(tmp);
3055 res = kern_path_create(dfd, tmp, path, is_dir);
3059 EXPORT_SYMBOL(user_path_create);
3061 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3063 int error = may_create(dir, dentry);
3068 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3071 if (!dir->i_op->mknod)
3074 error = devcgroup_inode_mknod(mode, dev);
3078 error = security_inode_mknod(dir, dentry, mode, dev);
3082 error = dir->i_op->mknod(dir, dentry, mode, dev);
3084 fsnotify_create(dir, dentry);
3088 static int may_mknod(umode_t mode)
3090 switch (mode & S_IFMT) {
3096 case 0: /* zero mode translates to S_IFREG */
3105 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3108 struct dentry *dentry;
3112 error = may_mknod(mode);
3116 dentry = user_path_create(dfd, filename, &path, 0);
3118 return PTR_ERR(dentry);
3120 if (!IS_POSIXACL(path.dentry->d_inode))
3121 mode &= ~current_umask();
3122 error = security_path_mknod(&path, dentry, mode, dev);
3125 switch (mode & S_IFMT) {
3126 case 0: case S_IFREG:
3127 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3129 case S_IFCHR: case S_IFBLK:
3130 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3131 new_decode_dev(dev));
3133 case S_IFIFO: case S_IFSOCK:
3134 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3138 done_path_create(&path, dentry);
3142 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3144 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3147 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3149 int error = may_create(dir, dentry);
3150 unsigned max_links = dir->i_sb->s_max_links;
3155 if (!dir->i_op->mkdir)
3158 mode &= (S_IRWXUGO|S_ISVTX);
3159 error = security_inode_mkdir(dir, dentry, mode);
3163 if (max_links && dir->i_nlink >= max_links)
3166 error = dir->i_op->mkdir(dir, dentry, mode);
3168 fsnotify_mkdir(dir, dentry);
3172 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3174 struct dentry *dentry;
3178 dentry = user_path_create(dfd, pathname, &path, 1);
3180 return PTR_ERR(dentry);
3182 if (!IS_POSIXACL(path.dentry->d_inode))
3183 mode &= ~current_umask();
3184 error = security_path_mkdir(&path, dentry, mode);
3186 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3187 done_path_create(&path, dentry);
3191 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3193 return sys_mkdirat(AT_FDCWD, pathname, mode);
3197 * The dentry_unhash() helper will try to drop the dentry early: we
3198 * should have a usage count of 1 if we're the only user of this
3199 * dentry, and if that is true (possibly after pruning the dcache),
3200 * then we drop the dentry now.
3202 * A low-level filesystem can, if it choses, legally
3205 * if (!d_unhashed(dentry))
3208 * if it cannot handle the case of removing a directory
3209 * that is still in use by something else..
3211 void dentry_unhash(struct dentry *dentry)
3213 shrink_dcache_parent(dentry);
3214 spin_lock(&dentry->d_lock);
3215 if (dentry->d_count == 1)
3217 spin_unlock(&dentry->d_lock);
3220 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3222 int error = may_delete(dir, dentry, 1);
3227 if (!dir->i_op->rmdir)
3231 mutex_lock(&dentry->d_inode->i_mutex);
3234 if (d_mountpoint(dentry))
3237 error = security_inode_rmdir(dir, dentry);
3241 shrink_dcache_parent(dentry);
3242 error = dir->i_op->rmdir(dir, dentry);
3246 dentry->d_inode->i_flags |= S_DEAD;
3250 mutex_unlock(&dentry->d_inode->i_mutex);
3257 static long do_rmdir(int dfd, const char __user *pathname)
3261 struct dentry *dentry;
3262 struct nameidata nd;
3264 error = user_path_parent(dfd, pathname, &nd, &name);
3268 switch(nd.last_type) {
3280 nd.flags &= ~LOOKUP_PARENT;
3281 error = mnt_want_write(nd.path.mnt);
3285 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3286 dentry = lookup_hash(&nd);
3287 error = PTR_ERR(dentry);
3290 if (!dentry->d_inode) {
3294 error = security_path_rmdir(&nd.path, dentry);
3297 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3301 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3302 mnt_drop_write(nd.path.mnt);
3309 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3311 return do_rmdir(AT_FDCWD, pathname);
3314 int vfs_unlink(struct inode *dir, struct dentry *dentry)
3316 int error = may_delete(dir, dentry, 0);
3321 if (!dir->i_op->unlink)
3324 mutex_lock(&dentry->d_inode->i_mutex);
3325 if (d_mountpoint(dentry))
3328 error = security_inode_unlink(dir, dentry);
3330 error = dir->i_op->unlink(dir, dentry);
3335 mutex_unlock(&dentry->d_inode->i_mutex);
3337 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3338 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3339 fsnotify_link_count(dentry->d_inode);
3347 * Make sure that the actual truncation of the file will occur outside its
3348 * directory's i_mutex. Truncate can take a long time if there is a lot of
3349 * writeout happening, and we don't want to prevent access to the directory
3350 * while waiting on the I/O.
3352 static long do_unlinkat(int dfd, const char __user *pathname)
3356 struct dentry *dentry;
3357 struct nameidata nd;
3358 struct inode *inode = NULL;
3360 error = user_path_parent(dfd, pathname, &nd, &name);
3365 if (nd.last_type != LAST_NORM)
3368 nd.flags &= ~LOOKUP_PARENT;
3369 error = mnt_want_write(nd.path.mnt);
3373 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3374 dentry = lookup_hash(&nd);
3375 error = PTR_ERR(dentry);
3376 if (!IS_ERR(dentry)) {
3377 /* Why not before? Because we want correct error value */
3378 if (nd.last.name[nd.last.len])
3380 inode = dentry->d_inode;
3384 error = security_path_unlink(&nd.path, dentry);
3387 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
3391 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3393 iput(inode); /* truncate the inode here */
3394 mnt_drop_write(nd.path.mnt);
3401 error = !dentry->d_inode ? -ENOENT :
3402 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3406 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3408 if ((flag & ~AT_REMOVEDIR) != 0)
3411 if (flag & AT_REMOVEDIR)
3412 return do_rmdir(dfd, pathname);
3414 return do_unlinkat(dfd, pathname);
3417 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3419 return do_unlinkat(AT_FDCWD, pathname);
3422 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3424 int error = may_create(dir, dentry);
3429 if (!dir->i_op->symlink)
3432 error = security_inode_symlink(dir, dentry, oldname);
3436 error = dir->i_op->symlink(dir, dentry, oldname);
3438 fsnotify_create(dir, dentry);
3442 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3443 int, newdfd, const char __user *, newname)
3447 struct dentry *dentry;
3450 from = getname(oldname);
3452 return PTR_ERR(from);
3454 dentry = user_path_create(newdfd, newname, &path, 0);
3455 error = PTR_ERR(dentry);
3459 error = security_path_symlink(&path, dentry, from);
3461 error = vfs_symlink(path.dentry->d_inode, dentry, from);
3462 done_path_create(&path, dentry);
3468 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3470 return sys_symlinkat(oldname, AT_FDCWD, newname);
3473 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3475 struct inode *inode = old_dentry->d_inode;
3476 unsigned max_links = dir->i_sb->s_max_links;
3482 error = may_create(dir, new_dentry);
3486 if (dir->i_sb != inode->i_sb)
3490 * A link to an append-only or immutable file cannot be created.
3492 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3494 if (!dir->i_op->link)
3496 if (S_ISDIR(inode->i_mode))
3499 error = security_inode_link(old_dentry, dir, new_dentry);
3503 mutex_lock(&inode->i_mutex);
3504 /* Make sure we don't allow creating hardlink to an unlinked file */
3505 if (inode->i_nlink == 0)
3507 else if (max_links && inode->i_nlink >= max_links)
3510 error = dir->i_op->link(old_dentry, dir, new_dentry);
3511 mutex_unlock(&inode->i_mutex);
3513 fsnotify_link(dir, inode, new_dentry);
3518 * Hardlinks are often used in delicate situations. We avoid
3519 * security-related surprises by not following symlinks on the
3522 * We don't follow them on the oldname either to be compatible
3523 * with linux 2.0, and to avoid hard-linking to directories
3524 * and other special files. --ADM
3526 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3527 int, newdfd, const char __user *, newname, int, flags)
3529 struct dentry *new_dentry;
3530 struct path old_path, new_path;
3534 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3537 * To use null names we require CAP_DAC_READ_SEARCH
3538 * This ensures that not everyone will be able to create
3539 * handlink using the passed filedescriptor.
3541 if (flags & AT_EMPTY_PATH) {
3542 if (!capable(CAP_DAC_READ_SEARCH))
3547 if (flags & AT_SYMLINK_FOLLOW)
3548 how |= LOOKUP_FOLLOW;
3550 error = user_path_at(olddfd, oldname, how, &old_path);
3554 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
3555 error = PTR_ERR(new_dentry);
3556 if (IS_ERR(new_dentry))
3560 if (old_path.mnt != new_path.mnt)
3562 error = may_linkat(&old_path);
3563 if (unlikely(error))
3565 error = security_path_link(old_path.dentry, &new_path, new_dentry);
3568 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
3570 done_path_create(&new_path, new_dentry);
3572 path_put(&old_path);
3577 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3579 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3583 * The worst of all namespace operations - renaming directory. "Perverted"
3584 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3586 * a) we can get into loop creation. Check is done in is_subdir().
3587 * b) race potential - two innocent renames can create a loop together.
3588 * That's where 4.4 screws up. Current fix: serialization on
3589 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3591 * c) we have to lock _three_ objects - parents and victim (if it exists).
3592 * And that - after we got ->i_mutex on parents (until then we don't know
3593 * whether the target exists). Solution: try to be smart with locking
3594 * order for inodes. We rely on the fact that tree topology may change
3595 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3596 * move will be locked. Thus we can rank directories by the tree
3597 * (ancestors first) and rank all non-directories after them.
3598 * That works since everybody except rename does "lock parent, lookup,
3599 * lock child" and rename is under ->s_vfs_rename_mutex.
3600 * HOWEVER, it relies on the assumption that any object with ->lookup()
3601 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3602 * we'd better make sure that there's no link(2) for them.
3603 * d) conversion from fhandle to dentry may come in the wrong moment - when
3604 * we are removing the target. Solution: we will have to grab ->i_mutex
3605 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3606 * ->i_mutex on parents, which works but leads to some truly excessive
3609 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3610 struct inode *new_dir, struct dentry *new_dentry)
3613 struct inode *target = new_dentry->d_inode;
3614 unsigned max_links = new_dir->i_sb->s_max_links;
3617 * If we are going to change the parent - check write permissions,
3618 * we'll need to flip '..'.
3620 if (new_dir != old_dir) {
3621 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3626 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3632 mutex_lock(&target->i_mutex);
3635 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3639 if (max_links && !target && new_dir != old_dir &&
3640 new_dir->i_nlink >= max_links)
3644 shrink_dcache_parent(new_dentry);
3645 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3650 target->i_flags |= S_DEAD;
3651 dont_mount(new_dentry);
3655 mutex_unlock(&target->i_mutex);
3658 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3659 d_move(old_dentry,new_dentry);
3663 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3664 struct inode *new_dir, struct dentry *new_dentry)
3666 struct inode *target = new_dentry->d_inode;
3669 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3675 mutex_lock(&target->i_mutex);
3678 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3681 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3686 dont_mount(new_dentry);
3687 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3688 d_move(old_dentry, new_dentry);
3691 mutex_unlock(&target->i_mutex);
3696 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3697 struct inode *new_dir, struct dentry *new_dentry)
3700 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3701 const unsigned char *old_name;
3703 if (old_dentry->d_inode == new_dentry->d_inode)
3706 error = may_delete(old_dir, old_dentry, is_dir);
3710 if (!new_dentry->d_inode)
3711 error = may_create(new_dir, new_dentry);
3713 error = may_delete(new_dir, new_dentry, is_dir);
3717 if (!old_dir->i_op->rename)
3720 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3723 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3725 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3727 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3728 new_dentry->d_inode, old_dentry);
3729 fsnotify_oldname_free(old_name);
3734 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3735 int, newdfd, const char __user *, newname)
3737 struct dentry *old_dir, *new_dir;
3738 struct dentry *old_dentry, *new_dentry;
3739 struct dentry *trap;
3740 struct nameidata oldnd, newnd;
3745 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3749 error = user_path_parent(newdfd, newname, &newnd, &to);
3754 if (oldnd.path.mnt != newnd.path.mnt)
3757 old_dir = oldnd.path.dentry;
3759 if (oldnd.last_type != LAST_NORM)
3762 new_dir = newnd.path.dentry;
3763 if (newnd.last_type != LAST_NORM)
3766 error = mnt_want_write(oldnd.path.mnt);
3770 oldnd.flags &= ~LOOKUP_PARENT;
3771 newnd.flags &= ~LOOKUP_PARENT;
3772 newnd.flags |= LOOKUP_RENAME_TARGET;
3774 trap = lock_rename(new_dir, old_dir);
3776 old_dentry = lookup_hash(&oldnd);
3777 error = PTR_ERR(old_dentry);
3778 if (IS_ERR(old_dentry))
3780 /* source must exist */
3782 if (!old_dentry->d_inode)
3784 /* unless the source is a directory trailing slashes give -ENOTDIR */
3785 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3787 if (oldnd.last.name[oldnd.last.len])
3789 if (newnd.last.name[newnd.last.len])
3792 /* source should not be ancestor of target */
3794 if (old_dentry == trap)
3796 new_dentry = lookup_hash(&newnd);
3797 error = PTR_ERR(new_dentry);
3798 if (IS_ERR(new_dentry))
3800 /* target should not be an ancestor of source */
3802 if (new_dentry == trap)
3805 error = security_path_rename(&oldnd.path, old_dentry,
3806 &newnd.path, new_dentry);
3809 error = vfs_rename(old_dir->d_inode, old_dentry,
3810 new_dir->d_inode, new_dentry);
3816 unlock_rename(new_dir, old_dir);
3817 mnt_drop_write(oldnd.path.mnt);
3819 path_put(&newnd.path);
3822 path_put(&oldnd.path);
3828 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3830 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3833 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3837 len = PTR_ERR(link);
3842 if (len > (unsigned) buflen)
3844 if (copy_to_user(buffer, link, len))
3851 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3852 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3853 * using) it for any given inode is up to filesystem.
3855 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3857 struct nameidata nd;
3862 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3864 return PTR_ERR(cookie);
3866 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3867 if (dentry->d_inode->i_op->put_link)
3868 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3872 int vfs_follow_link(struct nameidata *nd, const char *link)
3874 return __vfs_follow_link(nd, link);
3877 /* get the link contents into pagecache */
3878 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3882 struct address_space *mapping = dentry->d_inode->i_mapping;
3883 page = read_mapping_page(mapping, 0, NULL);
3888 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3892 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3894 struct page *page = NULL;
3895 char *s = page_getlink(dentry, &page);
3896 int res = vfs_readlink(dentry,buffer,buflen,s);
3899 page_cache_release(page);
3904 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3906 struct page *page = NULL;
3907 nd_set_link(nd, page_getlink(dentry, &page));
3911 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3913 struct page *page = cookie;
3917 page_cache_release(page);
3922 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3924 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3926 struct address_space *mapping = inode->i_mapping;
3931 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3933 flags |= AOP_FLAG_NOFS;
3936 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3937 flags, &page, &fsdata);
3941 kaddr = kmap_atomic(page);
3942 memcpy(kaddr, symname, len-1);
3943 kunmap_atomic(kaddr);
3945 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3952 mark_inode_dirty(inode);
3958 int page_symlink(struct inode *inode, const char *symname, int len)
3960 return __page_symlink(inode, symname, len,
3961 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3964 const struct inode_operations page_symlink_inode_operations = {
3965 .readlink = generic_readlink,
3966 .follow_link = page_follow_link_light,
3967 .put_link = page_put_link,
3970 EXPORT_SYMBOL(user_path_at);
3971 EXPORT_SYMBOL(follow_down_one);
3972 EXPORT_SYMBOL(follow_down);
3973 EXPORT_SYMBOL(follow_up);
3974 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3975 EXPORT_SYMBOL(getname);
3976 EXPORT_SYMBOL(lock_rename);
3977 EXPORT_SYMBOL(lookup_one_len);
3978 EXPORT_SYMBOL(page_follow_link_light);
3979 EXPORT_SYMBOL(page_put_link);
3980 EXPORT_SYMBOL(page_readlink);
3981 EXPORT_SYMBOL(__page_symlink);
3982 EXPORT_SYMBOL(page_symlink);
3983 EXPORT_SYMBOL(page_symlink_inode_operations);
3984 EXPORT_SYMBOL(kern_path);
3985 EXPORT_SYMBOL(vfs_path_lookup);
3986 EXPORT_SYMBOL(inode_permission);
3987 EXPORT_SYMBOL(unlock_rename);
3988 EXPORT_SYMBOL(vfs_create);
3989 EXPORT_SYMBOL(vfs_follow_link);
3990 EXPORT_SYMBOL(vfs_link);
3991 EXPORT_SYMBOL(vfs_mkdir);
3992 EXPORT_SYMBOL(vfs_mknod);
3993 EXPORT_SYMBOL(generic_permission);
3994 EXPORT_SYMBOL(vfs_readlink);
3995 EXPORT_SYMBOL(vfs_rename);
3996 EXPORT_SYMBOL(vfs_rmdir);
3997 EXPORT_SYMBOL(vfs_symlink);
3998 EXPORT_SYMBOL(vfs_unlink);
3999 EXPORT_SYMBOL(dentry_unhash);
4000 EXPORT_SYMBOL(generic_readlink);